ReAiLity Labs (@CreaitiveLabs) profile picture

Father, Enthusiast, Creator, Hobbyist My stack: 4090 RTX 64 GB RAM

Joined August 2025
Photo by ReAiLity Labs (@CreaitiveLabs)Photo by ReAiLity Labs (@CreaitiveLabs)Photo by ReAiLity Labs (@CreaitiveLabs)Photo by ReAiLity Labs (@CreaitiveLabs)Photo by ReAiLity Labs (@CreaitiveLabs)Photo by ReAiLity Labs (@CreaitiveLabs)Photo by ReAiLity Labs (@CreaitiveLabs)Photo by ReAiLity Labs (@CreaitiveLabs)Photo by ReAiLity Labs (@CreaitiveLabs)Photo by ReAiLity Labs (@CreaitiveLabs)
You don't need a $3,000 GPU to start making AI videos. Comfy Cloud gives you 5 free generations. I made a beginner guide showing exactly how to go from: Zero → Comfy Cloud → MiniMax H3 → reference image → AI video It takes about 5 minutes to follow. I documented every click so even if you've never touched ComfyUI before, you can do it. 📌 Bookmark this. 🔁 Repost it for someone getting started with AI video. Free tutorial in Comments
4
2
10
3,342
ReAiLity Labs retweeted
looking for a job Rt if you can 🙏
1
98
1
386
8,717
MiniMax H3 Image tip: A 2 MP image took 13.5s. Same prompt, new seed: 2.3s. The 6-step sampler is tiny. The 32B text encoder is the real cost, and ComfyUI caches it. So lock your prompt and roll seeds. All 4 models benchmarked: teapawt.net/CreaitiveLabs/status/2…
MiniMax H3 Image vs P-H3, Qwen 2.1 & MOSS: 80 images, one RTX 4090. H3 makes 4 MP images in 11.5s, but 80% of that is the text encoder. Change seeds, not prompts. Which model to use for what 👇
1
1
108
I made a song with AI, then tested Fable on it. My entire prompt: "Visualize this song using my open models in comfyui" This is what it built. I didn't give it a concept, a storyboard or a workflow. Details in the comments 👇👇👇
2
2
147
2/ What it did with that one sentence: • found ComfyUI running on my machine • checked which models I had installed • read the lyrics out of the song file • decided the song was about a freediver on one breath I found out the concept when I saw the first frames. 3/ Then it just kept going: • wrote a 22-shot storyboard locked to the song's bars • generated 28 stills (MiniMax H3 Image) • rendered 22 shots between them (Pika H3 Flex) • wrote its own editing script and cut the video All on my own RTX 4090. About an hour and a quarter of rendering. 4/ The part that got me: nobody asked for the timer. 5/ It also checked its own work. It looked at the stills, decided the opening frame was wrong (the diver appeared to be standing on the water), and redid it before rendering any motion. 6/ How much did I do? I typed one sentence and answered one multiple-choice question. The song is YuE 2. The video models are open and local. The direction, the pipeline and the edit are Fable 5.2 in Claude Code.
61
Models used: Claude Fable 5.2 in Claude Code: concept, storyboard, prompts, ComfyUI automation, edit. YuE 2: the song. MiniMax H3 Image: the stills. MiniMax H3 first/last-frame video model with the Pika H3 Flex LoRA: the motion. Qwen3-VL 32B: text encoder for both.
155
MiniMax H3 Image on one RTX 4090: 4 MP image in 11.5s 91% prompt adherence 46 GB on disk vs P-H3, Qwen 2.1 & MOSS, same 10 prompts, same seed. The 1 MP models are faster per image, but H3 makes 4× the pixels. Full breakdown: teapawt.net/CreaitiveLabs/status/2…
MiniMax H3 Image vs P-H3, Qwen 2.1 & MOSS: 80 images, one RTX 4090. H3 makes 4 MP images in 11.5s, but 80% of that is the text encoder. Change seeds, not prompts. Which model to use for what 👇
2
257
Lol...this will be reality soon
Are you f*cking kidding me. Got fired from OpenAI today. In my defense, being a $300k human API between Claude, ChatGPT, and Gemini worked perfectly until today...
Readers added context they thought people might want to know
This post features a fabricated termination email screenshot. @0xGoldTooth has no verifiable OpenAI employment history and posts AI-assisted "stories" to generate tools/content. github.com/0xGoldTooth x.com/0xGoldTooth/st… x.com/rfcwwide/statu…
48
ReAiLity Labs retweeted
Are you f*cking kidding me. Got fired from OpenAI today. In my defense, being a $300k human API between Claude, ChatGPT, and Gemini worked perfectly until today...
Readers added context they thought people might want to know
This post features a fabricated termination email screenshot. @0xGoldTooth has no verifiable OpenAI employment history and posts AI-assisted "stories" to generate tools/content. github.com/0xGoldTooth x.com/0xGoldTooth/st… x.com/rfcwwide/statu…
637
357
86
6,606
3,242,422
Same prompt. Same seed. 4 local models. "Elderly fisherman, white beard, rain on his face, 85mm" MiniMax H3 · P-H3 · Qwen 2.1 · MOSS Which one would you pick? 👇 Full benchmark (80 images, one RTX 4090): teapawt.net/CreaitiveLabs/status/2…
MiniMax H3 Image vs P-H3, Qwen 2.1 & MOSS: 80 images, one RTX 4090. H3 makes 4 MP images in 11.5s, but 80% of that is the text encoder. Change seeds, not prompts. Which model to use for what 👇
1
106
MiniMax H3 Image vs P-H3, Qwen 2.1 & MOSS: 80 images, one RTX 4090. H3 makes 4 MP images in 11.5s, but 80% of that is the text encoder. Change seeds, not prompts. Which model to use for what 👇
2
2
2
6
1,698
ReAiLity Labs retweeted
🚨 We're giving away a FULL YEAR of Claude Max to 2 of you! 🎁 Like + Repost + Reply to enter. Let's make it happen! 🚀 swarmpane.com/giveaway
264
275
4
378
23,144
ReAiLity Labs retweeted
We're giving away a Hogwarts Legacy custom wrapped NVIDIA GeForce RTX 5090 to celebrate its official release with DLSS 5 🪄 To enter: 🟢 Share this post 🟢 Comment #RTXON Terms and Conditions: nvidia.com/en-us/geforce/con…
84,766
72,208
7,710
67,739
6,011,627
Sunrise on the left. Aurora on the right. One sky. The final shot. ☀️🌌 Pika H3 Flex · 6 steps · one RTX 4090 🎬 Prompt in the comments ⤵️ Full film + workflow 👇 teapawt.net/CreaitiveLabs/status/2…
Pika H3 Flex in ComfyUI: 17 cinematic shots, ~96 s of footage, 50 minutes of render time on a single RTX 4090. Full breakdown, benchmark and every prompt inside.
2
4
68
Prompt ⤵️ Set the image below as the first frame, then paste: For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced. integrated_multimodal_description: [Shot 1] Live-action, cinematic, surreal epic with a teal and deep-blue grade, warm amber highlights and soft 35mm film grain, the shot begins from <Picture 1>: the snowy summit of the tallest mountain on Earth above an endless sea of clouds, the sky split between a golden sunrise on the left and a starry night with green aurora on the right, a tiny woman in a silver-blue gown on the summit with her silk cape unfurled like a banner. The sunrise and the aurora both shimmer across the split sky while her cape unfurls in the wind like a banner. The camera arcs slowly around the summit with large amplitude. overall_soundscape: A high, steady wind across the summit, with the cape fluttering.
1
40
The entire ocean, lifting into the sky toward the moon. 🌕 Pika H3 Flex · 6 steps · one RTX 4090 🎬 Prompt in the comments ⤵️ Full film + workflow 👇 teapawt.net/CreaitiveLabs/status/2…
Pika H3 Flex in ComfyUI: 17 cinematic shots, ~96 s of footage, 50 minutes of render time on a single RTX 4090. Full breakdown, benchmark and every prompt inside.
1
35
Prompt ⤵️ Set the image below as the first frame, then paste: For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced. integrated_multimodal_description: [Shot 1] Live-action, cinematic, surreal epic with a muted teal and deep-blue grade, silver highlights and soft 35mm film grain, the shot begins from <Picture 1>: a low-angle view from a black rock headland at night as the entire ocean lifts into the sky in a vast spiral of glittering water toward a huge full moon, a woman in a silver-blue gown at the edge watching, her silk cape rising in the wind. The ocean slowly lifts into the sky in a vast spiral of glittering water toward the moon while her cape rises in the wind. The camera tilts up following the spiral at slow speed. overall_soundscape: A deep, rushing roar of lifting water and swirling wind.
23